Skillv4.0.3

ClawScan security

Agent Architecture Guide · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignMar 13, 2026, 5:46 AM

Verdict: Benign
Confidence: high
Model: gpt-5-mini
Summary: This is an instruction-only guide about agent architecture patterns; its recommendations (writing working buffers, WAL, cron jitter, heartbeat batching, etc.) are coherent with the stated purpose and do not request unrelated credentials or install arbitrary code.
Guidance: This guide is coherent and useful for improving agent reliability, but it instructs agents to persist conversation-derived data and to call external endpoints. Before enabling automated writes or following the patterns in production, review where working buffers and WAL files will be stored, avoid logging secrets or sensitive personal data, confirm retention and access controls for those files, and verify any network calls (e.g., to clawhub.ai) are acceptable for your environment. If you plan to adopt these patterns, test them in a non-sensitive workspace first and audit written files regularly.

Review Dimensions

Purpose & Capability: okThe name and description match the SKILL.md content: patterns for agent reliability. Nothing in the skill asks for unrelated cloud credentials, exotic binaries, or system-wide config access — the guidance stays focused on agent architecture and operational practices.
Instruction Scope: noteThe instructions explicitly tell an agent to write persistent files (WAL, memory/working-buffer.md, TOOLS.md), run local CLI examples (openclaw cron), and call external endpoints (curl to clawhub.ai). These behaviors are aligned with the guide's purpose, but they do involve persistent logging of conversation context and network access — so users should be aware of privacy and data-retention implications before enabling automated writes or network calls.
Install Mechanism: okNo install spec or code files are present. Because this is instruction-only, nothing will be downloaded or written by an installer step — lowest install risk.
Credentials: okThe skill does not request environment variables, credentials, or config paths. Some examples show documenting required env vars in TOOLS.md, but the skill itself does not require any secrets or external credentials.
Persistence & Privilege: okalways is false and the skill is user-invocable. The guide recommends creating persistent files as an architectural choice, but it does not demand elevated platform privileges or global config changes. Autonomous invocation is allowed by default on the platform but is not a unique property of this skill.